TrackDiffusion icon indicating copy to clipboard operation
TrackDiffusion copied to clipboard
verified publisher icon pixeli99

May I ask how many iterations it takes for Stage 1 and Stage 2 respectively to converge on YouTube-VIS?

Open 12shuai opened this issue 10 months ago • 4 comments

12shuai avatar Feb 23 '25 08:02 12shuai

By the way, is it possible to train the model without Stage 1? Is the training able to converge?

12shuai avatar Feb 23 '25 08:02 12shuai

moreover, i find that the Instance-Enhancer is missing in the training&inference codes. is it necessary for SVD-version?

12shuai avatar Feb 23 '25 10:02 12shuai

Hi, @12shuai I want to confirm whether your basemodel is SVD or modelscope's T2V?

pixeli99 avatar Feb 24 '25 12:02 pixeli99

SVD

12shuai avatar Feb 24 '25 15:02 12shuai